Adaptive hash index partitions not tested in MTR

[Laurynas Biveinis]

Adaptive hash index partitions do not have a single MTR test. A basic test would be setting partitions to > 1 and running some workload with number of index trees >> AHI partitions and buffer pool size setting that would allow non-trivial amount of AHI pages. A bit more advanced test would be collecting failures, if any, from --mysqld=--innodb-adaptive-hash-index-partitions=N run to a testcase.

The bug does not apply for 5.6 but such testcases need to be created for https://blueprints.launchpad.net/percona-server/+spec/ahi-part-5-6 blueprint work.

[Laurynas Biveinis (laurynas-biveinis)]

MTR --mysqld=--innodb-adaptive-hash-index-partitions=8 run is clean: http://jenkins.percona.com/job/percona-server-5.5-param/824/

[Raghavendra D Prabhu (raghavendra-prabhu)]

Does this also test (innodb_buffer_pool_instances > 1) with
(innodb_adaptive_hash_index_partitions > 1), since that
particular combination has had a few bug reports earlier.

It would also be good to include this in RQG's combination
testing.

[Laurynas Biveinis (laurynas-biveinis)]

Raghu -

Code review tells that the number of buffer pool instances should be completely orthogonal to the number of AHI partitions. Were there any bugs that were repeatable with ibpi > 1 and not repeatable with ibpi = 1 while the number of AHI partitions was kept constant at > 1?

Re. RQG, see PQA-58 for AHI partitions and I'm asking Roel the status of ibpi > 1 testing there.

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-3007